The effects of the correct and incorrect modeling, the number of observations and the vicarious reinforcements were analyzed. The subjects were the 4th and 5th graders at the primary school and models were female undergraduate students. The task was the discriminative learning of heights. The results showed that the correct modeling was more effective than the incorrect modeling. However, the effectiveness of the incorrect modeling was unclear. As to the number of observations, in the correct modeling eight times was enough to achieve the asymptote, while in the incorrect modeling 16 times was not enough. It was found that the vicarious reinforcements was necessary for observational learning without which the model responses were avoided.